Inflation adaptation: how reserving actuaries are changing tack

Anomalously high levels of inflation were a global hot topic in 2022 and are expected to persist into 2023. As a consequence, many general insurance actuaries revisited how they allow for inflation in their modelling of claims reserves; we reflect on a range of approaches observed in different parts of the world.

Europe and United Kingdom

Historically, for most classes of business, inflationary allowances were included within the best estimate through:

• Standard Chain Ladder (CL) – implicitly, whereby historical inflation in the data and captured within development factors is assumed to continue to apply in the future.
• Initial Expected Loss Ratio (IELR) and Bornhuetter-Ferguson (BF) methods – implicit where the IELR is based on a historical averages and applied to an appropriately rate-adjusted premium. An explicit allowance is often made by on-leveling historical IELRs using a selected claims inflation index.

Frequency/Severity methods are less common, but are popular for Motor lines, where high inflation has already been an issue for many years. In these methods, the allowance is usually explicit, where the average cost per claim assumption is inflation adjusted.

Specific inflationary methods (such as inflation-adjusted CL) have seen limited use, however given the recent inflationary environment there is a push to use explicit inflationary loads that are explainable and justifiable. Such approaches include:

• Additional uplift of future cashflows using a forecast for future claims inflation on the class, reduced by the implicit inflation rate already assumed in the historical triangle.
• Additional uplift of development factors for selected years where higher than normal inflation is expected.
• Use of claims inflation forecast to on-level IELRs.

The theory is well understood in terms of selecting an appropriate inflation index for each class. However, a lack of resources has hindered teams in fully investigating the underlying drivers of inflation for each class. In practice , benchmark inflation measures (such as CPI) are selected and weighted based on the nature of the class, with an additional loading sometimes added to reflect inflation above the benchmark measure.

In some cases, high-level approaches are taken, such as an explicit inflationary load to the reserves (similar to how Solvency II ‘Events Not In Data’ loadings are applied). These may be informed by a series of stress/scenario tests on the underlying inflation assumptions.

Other considerations include:

• the extent to which model inputs (e.g. case reserves, premiums linked to sum insured) may already account for future inflation.
• the gearing effect for reinsurance arrangements (particularly non-proportional).

North America

Implicit allowance for inflation has also been common practice in North America, with the reliance that future inflation will not materially deviate from historical inflation.

Common approaches to allowing for the recent elevated level of inflation include:

• A ‘marginal’ allowance for explicit inflation on top of the implicitly-projected level. This involves applying loading by future calendar period and is most suitable where historical claims inflation has been stable but is expected to be markedly different to future inflation.
• Inflation-explicit reserving methods, which adjust historical development triangles to current price levels (e.g. inflation-adjusted Chain Ladder and Fischer-Lange methods) using an analysis of past inflation.

A key area of complexity has been the forecasting of cover-specific future inflation rates. With claims inflation commonly thought of as a combination of economic and social inflation, it is the significance of the latter (relative to other markets) that is proving more of a challenge. Establishing an allowance for social inflation requires significant judgement, with consideration given to how it will change as economic inflation changes. On the other hand, economic inflation can refer to published price level indices and their forecasts (e.g. CPI overall, CPI components such as medical care, ECI, PPI), weighted based on the type of coverage. Allowance for future inflation may also be considered stochastically using ESG-based projections.

As in other markets, consideration is given to the extent to which data (e.g. exposure measures) and assumptions already incorporate the appropriate level of inflation.

Australia

For long-tail classes, explicit allowance for inflation has been standard practice for many years. For traditional aggregate (triangle) reserving, economic and superimposed inflation are considered separately as follows:

• Economic inflation is typically allowed for by adjusting historical payment and case estimate triangles using an appropriate published index (e.g. Consumer Price Index, Wage Price Index) to present them as triangles of ‘current values’ for use in standard projection methodologies. Future economic inflation is determined with reference to economic forecasts and/or an inflation study.
• Superimposed inflation is typically allowed for by analysing average claim size data (adjusted for past economic inflation) by calendar period to identify historical trends for additional inflation. Together with expert judgement, this informs the selection of a future superimposed inflation assumption which is appropriate to each class or payment type.

Projected future cashflows are then inflated using the sum of future economic and superimposed inflation.

More advanced approaches to the measurement and application of inflation exist, particularly for granular (claims/policy-level) reserving methodologies and for products such as Lenders Mortgage Insurance (where macroeconomic variables are key drivers of claims behaviour).

Considerations which are common with other markets include:

• The extent to which inflation is allowed for in case estimates, and the impact on methodologies using incurred/case estimates (e.g. incurred Chain Ladder and Project Case Estimates).
• Impact on the full distribution of reserves, and the risk margin/adjustment in particular.

South Africa

Historically, in the South African market, inflation has not been considered explicitly in the reserving process. However, due to global supply chain issues, South African claims have been quite heavily impacted by inflation since Q3 2021. South African insurers have applied a variety of approaches to assess and adjust for the inflationary environment:

• Inflation-explicit methodologies such as inflation-adjusted chain ladder and average cost per claim have been considered to better understand the effect of the elevated inflation. These methods are mostly performed on large, material classes as it’s not the standard methodology and there are many data constraints. In such cases, appropriate assumptions / indices are required to estimate past and future claims inflation, depending on the underlying class of business.  Some insurers have found that the more data intensive approaches don’t necessarily lead to more informative results.
• Simpler methods such as adjustments to IELRs and BF methods are the more popular approach. Considerations when adjusting IELRs include:
  • Separate analysis of inflation on claim payments and expenses
  • Allowance for premium rate increases in response to inflation
  • Consideration of the impact on reinsurance recovery assumptions, where claim sizes continue to increase but attachment points remain unchanged.

Some insurers have also considered triangles constructed at more granular time periods (e.g. quarterly-quarterly instead of annual-quarterly). This enables trends by quarter to be more clearly identified and allowed for.

Conclusion

While inflation implicit, implicit-adjusted and explicit reserving methods are common to the general insurance markets described above, their prevalence and precise application varies greatly. This is typically driven by market characteristics such as regulatory requirements, norms around reserving methodologies and data availability. The recent paradigm shift to a higher inflationary environment is driving more of an adaptation of existing methodologies rather than fundamental changes in modelling approaches.